What is validation data used for?

• A. Generating synthetic data
• B. Verifying hardware compatibility
• C. Archiving outdated data
• D. Tuning model parameters during development

Answer: (B)

Validation data provides feedback during model development by offering a separate set of examples the model hasn’t trained on. This feedback is used to tune hyperparameters (like learning rate and regularization), adjust the model’s size or architecture, and decide when to stop training (early stopping). The aim is to improve generalization to new, unseen data, not just perform well on training data. After tuning, you reserve test data for the final evaluation. The other options aren’t what validation data is for: generating synthetic data is a different process to augment training data; verifying hardware compatibility falls under QA or system testing; archiving outdated data is a data-management task.